A Top-Down Framework to Autogenerate Mixed Integer Linear Programs for Supply Chain System Design
Abstract
This project will create a top-down modeling framework to recommend, evaluate, and refine supply chain designs without requiring lar,ge amounts of data nor specialized optimization modeling skills. The framework will consist of three layers. First, the input layer,will ask users targeted questions and to upload data in a specified format. The optimization layer will ingest this information and,algorithmically autogenerate a mixed integer linear program (MILP) that captures the functional relationships of the inputted data,,as well as supply chain structural rules and relationships. The generated MILP will be fed with inputted data and solved to recommen,d supply chain design decisions and to provide systematic performance estimates. The refinement layer will query users for feedback,on the recommended design. Based on the feedback, a refinement process will either adjust the model formulation, ask for additional, information from the user, or stop and provide the refined solution. Whilea tremendous amount of supply chain research creates math,ematical abstractions for supply chain design and analysis, existing approaches are a human-intensive process requiring substantial,time and optimization expertise to formulate a specific purpose-driven model. In contrast to existing bottom-up approaches, this pro,ject will create a top-down design framework to quickly recommend, evaluate, and refine supply chain designs for diverse contexts. M,ore information and/or autogenerating a higher fidelity mathematical model that captures more details of the specific context may pr,ovide a better design recommendation and more refined estimates but comes with model tractability challenges and may limit use of th,e framework. Thus, the focus of the proposed basic research effort will be to generate new knowledge to understand: In what design e,nvironments can tractable mixed integer linear programming models be autogenerated to provide useful supply chain designs and associ,ated estimates of system readiness and expected costs? New methods will be developed for each of the frameworks three layers by ans,wering the following open research questions:Input Layer: What are the set(s) of questions and input data structures that can achiev,e adequate expressive power with reduced user input request effort? Optimization Layer: How and at what level of model fidelity sh,ould different types of context-specific complexity be captured in the MILPs to generate tractable models and useful design recommen,dations? Refinement Layer: How often and in what ways does the framework need to query the user to converge to a final refined solu,tion? The proposed framework is intended for use at the design stage, where multiple configurations, policies, and use cases need to, be evaluated quickly by a diverse group of people, including users who may not be specialists in combinatorial optimization, and wh,ere existing solution data is limited. By codifying the expertise of a human optimization expert, if successful, the proposed top-do,wn framework can democratize and increase the use of optimization in Naval and commercial applications.Approved for Public Release
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Jul 13, 2022
- Source ID
- N000142212542
Entities
People
- Jennifer Pazour
Organizations
- Office of Naval Research
- Rensselaer Polytechnic Institute
- United States Navy